PLDI 2024
Mon 24 - Fri 28 June 2024 Copenhagen, Denmark
Mon 24 Jun 2024 10:55 - 11:10 at Iceland - Optimization Chair(s): Aviral Shrivastava

In this article, we present a new method for implementing a neural network whose weights are floating-point numbers on a fixed-point architecture. The originality of our approach is that fixed-point formats are computed by solving an integer optimization problem derived from the neural network model and concerning the accuracy of computations and results at each point of the network. Therefore, we can bound mathematically the error between the results returned by the floating-point and fixed-point versions of the network. In addition to a formal description of our method, we describe a prototype that implements it. Our tool accepts the most common neural network layers (fully connected, convolutional, max-pooling, etc.), uses an optimizing SMT solver to compute fixed-point formats and synthesizes fixed-point C code from the Tensorflow model of the network. Experimental results show that our tool is able to achieve performance while keeping the relative numerical error below the given tolerance threshold. Furthermore, the results show that our fixed-point synthesized neural networks consume less time and energy when considering a typical embedded platform using an STM32 Nucleo board.

Mon 24 Jun

Displayed time zone: Windhoek change

10:40 - 12:20
OptimizationLCTES at Iceland
Chair(s): Aviral Shrivastava Arizona State University
10:40
15m
Talk
Accelerating Shared Library Execution in a DBT
LCTES
Tom Spink University of St Andrews, Björn Franke University of Edinburgh
10:55
15m
Talk
Efficient Implementation of Neural Networks Usual Layers on Fixed-Point Architectures
LCTES
Dorra Ben Khalifa University of Toulouse - ENAC, Matthieu Martel Université de Perpignan Via Domitia
11:10
15m
Talk
TinySeg: Model Optimizing Framework for Image Segmentation on Tiny Embedded Systems
LCTES
Byungchul Chae Kyung Hee University, Jiae Kim Kyung Hee University, Seonyeong Heo Kyung Hee University
11:25
10m
Break
Break - 10 minutes
LCTES

11:35
15m
Talk
MixPert: Optimizing Mixed-Precision Floating-Point Emulation on GPU Integer Tensor Cores
LCTES
Zejia Lin Sun Yat-sen University, Aoyuan Sun Sun Yat-sen University, Xianwei Zhang Sun Yat-sen University, Yutong Lu Sun Yat-sen University
11:50
15m
Talk
Optimistic and Scalable Global Function Merging
LCTES
12:05
15m
Talk
(Invited paper) Language-Based Deployment Optimization for Random Forest
LCTES
Jannik Malcher TU Dortmund University, Daniel Biebert TU Dortmund University, Kuan-Hsun Chen University of Twente, Sebastian Buschjäger TU Dortmund University, Christian Hakert TU Dortmund University, Jian-Jia Chen TU Dortmund University